If everything a model writes for you sounds like everything it writes for everyone else, that is usually a prompt problem with a straightforward fix.

You asked for the average

A model given no constraints returns the middle of everything it has seen. Write a blog post about productivity has one obvious answer, and it is the one you got. The narrower the brief, the further from the average the output lands.

Three changes that work

• Give it your own material to work from. Real copy, real transcripts, real customer language.
• Ban the phrasing you keep seeing. Explicit prohibitions work better than asking for originality.
• Ask for several distinct angles rather than one answer, then say which angle you want developed.

The specificity test

Read the output and ask whether a competitor could publish it unchanged. If they could, the prompt did not contain anything that was distinctively yours. Fix the prompt, not the output.

Editing is not the answer

It is tempting to fix generic output by rewriting it. That works once, and then you do it again tomorrow. Improving the prompt fixes every future run, which is the whole point of keeping a prompt library rather than a folder of edited drafts.